AI firms could need approval before launching in the UK
Key Points
- High-risk AI systems would need prior approval before deployment in the UK
- Due diligence duties proposed for every stage of the AI supply chain
- Requirements would scale by supply chain role and severity of risk
- Regulators currently cannot test or block AI systems before release
- Low-risk systems would face lighter legal requirements under the risk-based model
Companies building high-risk AI systems could need government approval before launching them in the UK under proposals from the Joint Committee on Human Rights.
In a new report on human rights and the regulation of AI, the committee said AI systems that pose a high risk of causing human rights harms should require prior approval before deployment.
The proposal would mark a major shift from the current position, where regulators cannot test or evaluate AI systems before release and cannot block a launch on safety grounds.
The committee said developers currently engage with the AI Security Institute on a voluntary basis, and the institute has no statutory power over them.
Under the proposed AI Bill, obligations would apply at every stage of the AI lifecycle rather than only at the point of use. The committee said everyone in the supply chain responsible for designing, amending and using AI systems should know their obligations and meet them.
Legislation should require actors at all stages of the supply chain to carry out due diligence to prevent unacceptable risks to human rights when designing, developing or deploying AI, the report said. It said those requirements should scale according to a company’s role in the supply chain and the gravity of the risk involved.
The committee said UK laws that could apply to AI currently do so mainly at deployment, which places responsibility on users of the technology rather than the companies that design it.
“New legislation is needed to establish a comprehensive set of protections that deal with the entire AI supply chain and its lifecycle,” said Alex Sobel, chair of the Joint Committee on Human Rights.
The report said the new regime should classify AI systems by risk level, with low-risk systems facing fewer and less demanding requirements. The committee said the law should not place unjustified burdens on businesses, so UK companies can continue to innovate.
It also called for mandatory transparency requirements across the AI lifecycle, including an obligation for systems with significant impacts on people to state when and how they are being used.
The committee said governments need to collaborate internationally to identify AI risks and promote consistent standards, given the global nature of AI harms.
“It is moving with such speed and complexity that its impact is hard to accurately predict,” said Sobel. “What is clear is that at present we are unprepared to deal with its consequences however potentially dire they may be.”
